I asked AI to design a bass cab

[ihixulu]

Like any other tool, AI needs to be used properly. And like any other tool, there's something of learning curve to learn how to do that.

By the time you've curated the training data set so that it will give you a reasonable result you could have saved the effort (and all that electricity) and read a book or communicated to a human who knows what they're doing (god forbid we talk to each other...) and designed something reasonable.

 

[Koshchei]

Like any other tool, AI needs to be used properly. And like any other tool, there's something of learning curve to learn how to do that.

Not quite specific enough to be accurate. AI is best used for classification and data regression problems. While Anthropic has done some remarkable stuff with Claude, keep in mind that it's still data regression and classification that's driving the output, which is why it isn't quite (and never will be) a solution, even though sometimes it might be serviceable as one.

If we ever enter a sci-fi world where AGI comes to be, then obviously all of this is out the window, but I suspect that we'll have bigger problems at that point than having a giant electronic slave mind design speaker cabinets for people who don't trust the provided backline on fly-in gigs.

 

[lenuaim]

Can I join ? I don't really want to start a new thread, but since someone mentioned the 10FE200 I'll say that I've thought of a Faital powered cabinet these days. One cabinet, many heads, many uses. I modeled the 12FE300 and the new 12FE400 in the 2x12 cabinet I have planned (cca. 120 but realistically fewer litres), They come in 4 or 8 ohm and it seems they're in the same ballpark. I live in Europe and it's easier to get them here. The 12FE400 8ohm is cheapest but I don't know if its specs show such a big difference to the others. I frequently switch between amps, basses, flatwounds, roundwounds, (i do prefer heavy gauge flats tho) and i also play various kinds of music.
Lack of space limits me to just one cabinet to rule them all, one cabinet to find them.
Qtc for 12FE300 8ohm = 0.791
Qtc for 12FE400 4ohm = 0.765
Qtc for 12FE300 4ohm = 0.684 (what gives ?)

 

[Doctor Morbius]

I did too, but I forgot to specify that it needed to be in our dimension.

 

[agedhorse]

4 and 8 ohm versions of the "same" driver are often surprisingly different.

It looks like you are plotting the normalized response which is NOT representative of the important differences.

What happens when you plot the SPL with an input POWER of 1 watt? (not 2.83 volts) on all drivers???

 

[lenuaim]

Thanks @agedhorse for the reply; This happens with 1W. With a mighty 50W of power (my main amp is a Dynacord Bass King T with 2x6CA7 tubes from the late 60s that only pushes ~50ish watts) the difference in the SPL graphs is also <1dB. Just the cabinet, no lining, no filling, Qa=100. The odd one out is the 12FE330 which other sites, for example usspeaker, say it's for 2 or 3-way systems, studio monitors etc.
I know the differences are varied, sometimes wildly, between speakers. but this is what i'm seeing. Is there something i'm doing wrong in this plot ? I've even written an e-mail to FaitalPRO and really curious to see what they have to say. I also have a nice 12 watt Vermona MV3 tube head with a quirky 6ohm output. And this is really the main reason why I'm pondering on whether i should go for a 2x4ohm series or a stereo 8ohm thing (I have used it before on an 8ohm cabinet and it was all fine).

 

[fdeck]

I'm an avid DIY'er and "citizen scientist," and my non commercial web page includes an article on how speakers work, plus my own modeling program. It doesn't have any great advantages over WinISD, other than being fully open and documented.

Something I've learned is that "different" drivers can end up looking similar to one another in the response curves for a couple of reasons. First, the box becomes a dominant factor. The box modifies both the resonant frequency and the total Q factor. Second, the logarithmic function for computing dB turns big changes into small ones. Thus, speakers actually turn out to be forgiving of minor changes to the design parameters.

Germane to what Andy mentioned, most modeling programs, including mine, use the "power" number to compute an equivalent input voltage, and the graphs are shown for constant voltage. This approach has its own rationale, but is never clearly explained by the modeling apps. In fact a motivation for writing my own modeling program is that I don't want to trust a "black box" if I don't know what it's doing.

 

[agedhorse]

First, it's unusual for the 4 ohm version to have a higher sensitivity than the 8 ohm version (it's possible if the optimization was for 4 ohms but not common).

Second, are you using WinISD? If so, I have seen way too many incorrect results compared to known good tools to trust any of the results.

 

[fdeck]

One thing I know about WinISD (confirmed somewhere in their documentation) is that they compute the input voltage from the DC resistance, not from the nominal impedance. You can override this by entering an actual voltage. But if the 4- and 8-Ohm versions have something other than a 2:1 ratio of R_e values, it could give rise to an unexpected difference in the SPL curves.

In my own case, I care more about voltage than power because voltage is the "analog" of the input signal, and is what ultimately limits the output of my amplifiers.

Not to self-promote, but an advantage of "open source" is that if I'm wrong, someone can correct me and I can fix it. Or if one of my assumptions is controversial, then at least I can put up a warning.

For the OP, I'd also suggest checking out possible 1x15 solutions. Speaking strictly as a DIY'er, my experience has been that single driver designs tend to be more economical in terms of weight and cost, and slightly less complex to build.

 

[agedhorse]

Using nominal power and nominal impedance makes comparisons easier and more uniform overall.

 

[hbarcat]

I've yet to see one I'd want to build, but if enough of you guys keep training the models I expect they will get there in a few years or so. The trick, as always, is knowing the right questions to ask.

GIGO.